Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Modelling the Effect of Audience Tuning on Generative Episodic Memory

Zahra Fayyaz¹, Aya Altamimi², Ullrich Wagner³, sen Cheng⁴, Gerald Echterhoff, Laurenz Wiskott¹; ¹Ruhr-Universität Bochum, ²INI/RUB, Ruhr-Universtät Bochum, ³Westfälische Wilhelms-Universität Münster, ⁴Ruhr-Universtät Bochum

Presenter: Aya Altamimi

Episodic memory is highly malleable and shaped by social interactions.\ Rather than storing exact representations, it reconstructs incomplete traces using semantic information influenced by motivation and context. In shared reality situations, as shown by the saying-is-believing (SIB) paradigm, verbalizing information affects both audience perception and the speaker’s memory. Yet, most computational models assume faithful storage and neglect social influence. We have developed a generative episodic memory model combining a VQ-VAE for visual perception, a masking module for attention and hippocampal storage, and a transformer for semantic completion. Images are encoded, partially stored, and later reconstructed into plausible, not necessarily accurate, scenarios. The model introduces key innovations: blending multiple memory traces, incorporating emotional valence and biased reconstruction, and handling ambiguous stimuli. These features allow it to simulate SIB effects and social influences on memory, offering insights into how communication and context shape what and how we remember.

Topic Area: Memory, Spatial Cognition & Skill Learning

Extended Abstract: Full Text PDF